What happens when businesses pay for task execution instead of software?
Traditional Software as a Service (SaaS) has long been defined by subscriptions and licenses justified by users better completing their tasks. Whereas agentic AI is about managing teams of agents to complete those tasks for you. This is more computationally expensive so introduces business models that charge for usage, for tasks executed, not just subscriptions.
This appears to be a subtle shift but OpenAI has publicly floated the idea of intelligent agents priced at a wage-like $2,000/month, not so subtle. But if agents aren't people, do you owe them a wage? Surely business would pay for what ai agents achieve, not offer them a fixed income.
Are these changes the new path to business efficiency—or just a new headache? Let’s dive into how agents employ different business models to SaaS.
Comparing Business Models
SaaS | Agentic AI | |
Pricing Model | Subscription-based: predictable fixed costs, regardless of usage. | Subscription + Usage. Occasionally outcome-based. |
Cost Dynamics | Fixed costs dominate (subscriptions and wages). Budgeting and accounting focus on predictable expenses. | Variable costs dominate (pay per task or outcome). Accountants may view costs as directly tied to activities (Activity-Based Costing). |
Cost Variability | Low variability: predictable subscription fees, even if software is underused. | High variability: costs scale with usage, resembling "cloud-style" cost structures for activity-based pricing. |
Business Accounting | Fixed costs (subscriptions, wages) allow stable financial forecasting but unrelated to performance | Variable costs complicate forecasting but align with task-specific value (e.g., ABC accounting). |
Scalability | Scaling requires additional licenses or subscriptions, making growth incremental and less flexible. | Scaling is dynamic: costs grow directly with task demand, allowing real-time adjustments but introducing budgeting complexity. |
Business Performance Implications
SaaS | Agentic AI | |
Process Efficiency | Efficiency depends on user skill and ability to utilize software effectively. | AI automates task execution, driving efficiency by removing the dependency on user expertise for task completion. |
Quality Control | Users maintain control; quality depends on user configuration and input. | Agentic AI quality depends on: - data quality - workflow quality - prompt quality Disappointment may follow where users lack experience |
Innovation | Vendors focus on broad feature development and tools that appeal to general user bases. | Vendors focus on allowing users to adapt the data, workflow and prompts to their needs. |
Risk of Overuse | Low risk: subscriptions provide stable costs regardless of volume of use. | High risk: excessive or redundant task execution inflates costs, requiring careful monitoring and management. |
Show Me the Incentive and I’ll Show You the Outcome.
As Charlie Munger (Berkshire Hathaway) famously observed, “Show me the incentive and I’ll show you the outcome.” Agentic AI’s activity-based cost model incentivises efficiency and task-specific optimisation but introduces variability, occasionally disappointment, where the agent team need to rework incomplete tasks.
Imagine you are a small team in a big company, think of the hoops you must jump through to get budget for software licenses. How does budget control function when the majority of tasks are charged variably? As if everyone in the company was a contractor hired in the morning and paid by the hour.
We're in an exploratory phase, as we noted, OpenAI want you to put them on the payroll for $2000/mth, but is this waht business clients want? While SaaS subscriptions offered stability and broad functionality, agentic AI permits sharper focus on the value of individual tasks, potentially transforming how businesses deliver outcomes and define their strategies.
The question isn’t simply what these incentives will lead to —but also whether businesses are ready to manage the complexities.
Hidden Costs & Doughnut Jobs
We only want to pay for work well done; quality results from AI agents via quality data and workflows. We'll need to automate the process of evaluating the work so we know whether it ticks that box and payment can be authorised. 'We' means you too, we all become managers in the world of ai agent teams. Being a manager of agents makes us productive, but imposes a hidden cost.
That hidden cost is already common with coding agents. This is a great example with lessons for other professions. AI agents are very productive writing code, then will happily write faster than anyone can read, let alone understand. Hence the hidden cost :
The developer requires a new level of foresight for what to request.
Its easy to take the wrong path, asking for something you didn't need
or to be paralysed by not knowing what the priority should be
In other words, the developer shifts into being a development manager
And all that code will be published under the owner's name, so now we need to
Evaluating far more code than we could ever have written
This makes us a tester too, possibly not a job we applied for
Agentic coding with Codeium's Windsurf has already changed my work day fundamentally. I am spending more time planning my work and, when complete, testing and integrating it. The middle is increasingly missing, automated by my team. It's gone doughnut shaped!
Beware it takes time for organisations and their people to adapt to such new realities, so it will take time to realise the productivity gains of agents.
Fast, Skilled or Affordable - You Can Pick Only Two
Traditional management faced a simple trade-off when hiring people : "Fast, Skilled, Affordable - pick two."
Hire Quickly + Demand Skill = Not easily affordable (immediate hire of top talent)
Demand Skill + Affordability = Not fast (long recruiting for budget-friendly experts)
Hire Quickly + Demand Affordability = Not skilled (quick hire requiring training)
This rule of thumb does not apply to AI agents. They can be hired immediately, speed is no longer a constraint. With AI agents, staff will be faced with a new trilemma.
Staff must now act like managers and govern their automated teams. They must take time to describe the task, manage the workflow and evaluate the outputs. In other words, "Governance".
Agentic AI: Governed, Skilled, Affordable - Pick Two
As we discussed, management costs time and money. The greater the output the more management is required to instruct and evaluate that output.
Let's image we are hiring AI agents, our trilemma is those agents can be:
Well governed + Skilled = Not affordable (Oversight takes time, skills cost money)
Skilled + Affordable = Not well governed (Budget unavailable to manage outputs)
Well governed + Affordable = Not skilled (No budget for skills, or governance limits agent)
As AI grows more powerful, governance only gets harder - each new ability demands more sophisticated workflow design, management oversight and output evaluation. Tomorrow's winners won't just deploy AI, they'll master governing it.
Agents Earning for Themselves?
The obvious answer is to have AI agents as managers, but if they both manage and execute, then they are a business. This thought is radical, but it has already been done, see the story of Goatseus Maximus and the GOAT coin, now with market cap of $1bn (20-Nov-2024).
That's an idea for another day.
Agentico is among the first Agentic AI advisors in the UK, supported by 10 yrs in Machine Learning and 20yrs in analytics. Make sense of, and leverage, the seismic changes that AI agents are bringing to marketing or your industry. We're happy to talk more about the opportunities of Agentic AI and ML at your organization or event. Get in Touch Agentico.ai: AI with humans in mind |
Comments